Sber Proposes Creation of an AI Development Council: Experts Believe the Initiative Will Accelerate Technology Growth in Russia

Sber has proposed the creation of a Council for Generative Software Development under the patronage of the Ministry of Digital Development, as reported by Vedomosti. The platform would bring together IT companies, customers, and scientists to establish common rules for using neural networks in software development. Experts believe that businesses primarily need unified security requirements for code created by AI agents, followed by methods for evaluating their effectiveness. The timeline for the Council's creation is not yet known. According to Sber's proposal, the Council would define rules for AI integration into software development and evaluate the efficiency of generated code, potentially forming the basis for future national standards. Minister of Digital Development Maksut Shadayev positively assessed the initiative and proposed preparing concrete proposals by the end of the year in three areas: economic efficiency of generative development, use of computational resources, and development of common technological components. Sber is ready to share its experience in implementing AI among its 50,000 engineers. Discussions also involved representatives from Rostelecom, T-Bank, VTB, Rosatom, Russian Railways, and Astra Group. In generative development, an AI agent can write code, analyze codebases, test, and fix errors, while the programmer sets tasks and controls results. Roman Venediktov, CEO of Emergent Multi-Agent Systems LLC, noted that companies currently often use external AI models and rarely run their own internal models, forming their own regulations for admitting AI agents, which often slows down adoption. Experts point out that the most important task is to standardize security and evaluation methods for AI-generated code. Dmitry Kryukov, head of AI direction at MTS Link, believes that unified requirements for security and quality are crucial, followed by transparent evaluation methodologies. He also stressed that inventing a special 'responsibility for AI errors' regime is unnecessary, as AI agents are tools and the product's creator is responsible. However, experts caution that standards may lag behind AI development. Dmitry Pilipenko, deputy general director of IT holding LANSOFT, advises not to bind requirements to specific models or technologies, otherwise rules may become outdated. Dmitry Kryukov also suggests that standards should describe verification methods rather than specific technologies. Dmitry Yudin, head of AI direction at Cloud.ru, believes that clear rules could accelerate the adoption of AI agents, especially in small and medium businesses, by reducing costs and increasing trust, while also considering security and employee readiness.